> ## Documentation Index
> Fetch the complete documentation index at: https://docs.aieev.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Qwen3.8-27B
> Compact dense vision-language model for agentic and long-horizon work.
**Model ID:** `qwen/qwen3.8-27b` | **Endpoint ID:** `qwen-3-8-27b` | **Parameters:** 27B (dense) | **Released:** 2026-08-20
| Input Price | Output Price | Context | Modality | Quantization |
| :----------: | :----------: | :-----: | :-------------------------: | :----------: |
| \$0.400 / 1M | \$3.000 / 1M | 262,144 | text + image + video → text | BF16 |
## Overview
Built on the architectural foundation of Qwen3.5, Qwen3.8 delivers substantial gains across coding, professional work, research, and long-horizon agentic tasks. Qwen3.8-27B brings these advances to a compact, deployment-friendly dense model: a native vision-language model that understands images and videos, with flexible thinking control, designed to carry complex, multi-step tasks through to completion with greater reliability.
Try the model in the playground.
Deploy with AIR Container.
Connect with the OpenAI SDK as-is.
## Pricing
| Currency | Input (1M tokens) | Output (1M tokens) |
| :------- | ----------------: | -----------------: |
| USD | \$0.400 | \$3.000 |
| KRW | ₩580 | ₩4,350 |
## Supported Features
All features are available through the [OpenAI Compatible API](/docs/air-api/openai-compatible-api).
| Feature | Supported | Notes |
| :-------------------------------------------------------------------------------- | :-------: | :------------------------------------------ |
| [Streaming](/docs/air-api/openai-compatible-api#streaming) | ✅ | SSE, `include_usage` supported |
| [Tool Calling](/docs/air-api/openai-compatible-api#tool-calling-function-calling) | ✅ | Parallel and streaming tool calls supported |
| [Vision](/docs/air-api/openai-compatible-api#vision-image-input) | ✅ | Images and video (`image_url`, `video_url`) |
| [Reasoning](/docs/air-api/openai-compatible-api#reasoning) | ✅ | Depth is tunable with `reasoning_effort` |
| [JSON Mode](/docs/air-api/openai-compatible-api#json-mode-structured-outputs) | ✅ | `json_object` and `json_schema` |
| logprobs | ✅ | `logprobs` / `top_logprobs` |
When streaming, this model returns reasoning in the `reasoning_content` delta field rather than `reasoning`.
## Key Features
* Native vision-language model — understands images and video without a separate adapter
* Flexible thinking control: turn reasoning on or off, and tune its depth with `reasoning_effort`
* Reasoning from previous turns can be preserved to keep multi-step decisions consistent
* Long-context inference up to 262,144 tokens, with up to 131,072 output tokens
* Strong agentic coding and tool use across multi-step tasks
* Multilingual across 13 languages including Korean, English, Chinese, and Japanese
* Served in BF16 for full-precision quality
## Use Cases
### Agentic Coding
Work through a repository across multiple steps rather than answering a single question.
```text Input theme={null}
This Flask endpoint returns 500 under concurrent load.
Find the race condition and patch it.
```
```text Output theme={null}
The counter increment is not atomic, so two requests can read the same
value before either writes back. Wrapping the update in a lock, or moving
the counter to an atomic backend, closes the lost-update window.
Patch follows.
```
### Document and Chart Understanding
Read charts, tables, scanned documents, and technical diagrams from images.
```text Input theme={null}
[chart image] What does this throughput curve tell us?
```
```text Output theme={null}
Throughput peaks at concurrency 32 and declines beyond it, which means
the batch scheduler saturates before the memory bound is reached.
```
### Video Understanding
Summarize long-form video and answer questions about what happens in it.
## Parameters
| Parameter | Type | Required | Default | Description |
| -------------------- | ---------------- | -------- | ------- | ------------------------------------------------------------------------------------------------ |
| `messages` | array | Required | - | Chat messages. Content may be a string or an array of blocks (`text`, `image_url`, `video_url`). |
| `max_tokens` | integer | Optional | 32768 | Maximum output tokens, including reasoning tokens. |
| `temperature` | number | Optional | 1 | Sampling temperature. 1.0 in thinking mode, 0.7 in non-thinking mode. |
| `top_p` | number | Optional | 0.95 | Nucleus sampling. 0.95 thinking, 0.80 non-thinking. |
| `top_k` | integer | Optional | 20 | Top-k sampling. Set `-1` to disable. |
| `min_p` | number | Optional | 0 | Minimum probability threshold relative to the top token. |
| `presence_penalty` | number | Optional | 0 | Raise toward 2 to reduce endless repetition. |
| `frequency_penalty` | number | Optional | 0 | Supported but outside the recommended parameter set. |
| `repetition_penalty` | number | Optional | 1 | Passed through as an extra body field. |
| `reasoning_effort` | string | Optional | xhigh | Reasoning depth — `xhigh`, `medium`, or `low`. |
| `enable_thinking` | boolean | Optional | true | Emit a reasoning block before the answer. |
| `preserve_thinking` | boolean | Optional | true | Keep reasoning from previous turns for consistency. |
| `tools` | array | Optional | - | Tool definitions in OpenAI function-calling format. |
| `tool_choice` | string \| object | Optional | auto | Tool selection policy. |
| `response_format` | object | Optional | - | `json_object` for JSON mode, `json_schema` for structured outputs. |
| `stop` | array | Optional | - | Up to 4 strings that halt generation. Not applied inside the reasoning block. |
| `seed` | integer | Optional | - | Seed for reproducible sampling. |
| `logprobs` | boolean | Optional | false | Return log probabilities for output tokens. |
| `top_logprobs` | integer | Optional | - | Alternatives per position. Requires `logprobs`. |
| `stream` | boolean | Optional | false | Enable streaming. |
| `stream_options` | object | Optional | - | Set `include_usage` to receive token counts in the final chunk. |
## Model Details
| Property | Value |
| ------------------ | ------------------------------------------------------------------------------------------- |
| Context Length | 262,144 |
| Max Output Length | 131,072 |
| Quantization | bf16 |
| Input Modalities | text, image, video, file |
| Output Modalities | text |
| Supported Features | tools, reasoning, streaming, vision, json\_mode, structured\_outputs, logprobs, web\_search |
| Hugging Face ID | `Qwen/Qwen3.8-27B` |
## Quick Start
Generate an API key from your [AirCloud account](https://aieev.com).
Replace `YOUR_API_KEY` with your actual key and choose your preferred language (Python, cURL, Node.js).
```python Python theme={null}
from openai import OpenAI
client = OpenAI(
api_key="YOUR_API_KEY",
base_url="https://ap-1.aieev.cloud/endpoints/qwen-3-8-27b/v1"
)
response = client.chat.completions.create(
model="qwen/qwen3.8-27b",
messages=[{"role": "user", "content": "Hello!"}]
)
print(response.choices[0].message.content)
```
```bash cURL theme={null}
curl https://ap-1.aieev.cloud/endpoints/qwen-3-8-27b/v1/chat/completions \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "qwen/qwen3.8-27b",
"messages": [{"role": "user", "content": "Hello!"}]
}'
```
```javascript Node.js theme={null}
import OpenAI from "openai";
const client = new OpenAI({
apiKey: "YOUR_API_KEY",
baseURL: "https://ap-1.aieev.cloud/endpoints/qwen-3-8-27b/v1",
});
const response = await client.chat.completions.create({
model: "qwen/qwen3.8-27b",
messages: [{ role: "user", content: "Hello!" }],
});
console.log(response.choices[0].message.content);
```